SpeechBrain Lang Detect

An audio transformer plugin for OpenVoiceOS. It detects the spoken language of an audio signal using a SpeechBrain classifier, before speech-to-text runs.

Supported models

The plugin loads any of these SpeechBrain models from Hugging Face:

• speechbrain/lang-id-commonlanguage_ecapa - 45 languages
• speechbrain/lang-id-voxlingua107-ecapa - 107 languages
• TalTechNLP/voxlingua107-epaca-tdnn - 107 languages
• TalTechNLP/voxlingua107-epaca-tdnn-ce - 107 languages
• sahita/lang-VoxLingua107-ecapa - English, Hindi
• sahita/language-identification - English, Hindi, Other

Install

pip install ovos-audio-transformer-plugin-speechbrain-langdetect

Usage

Add the plugin to the audio_transformers section of your listener config. Set model to any of the SpeechBrain models listed above.

"listener": {
    "audio_transformers": {
        "ovos-audio-transformer-plugin-speechbrain-langdetect": {
            "model": "speechbrain/lang-id-commonlanguage_ecapa"
        }
    }
}

Set use_cuda to true to run the classifier on a CUDA GPU instead of the CPU.

The plugin only classifies the language when more than one language is listed as valid. If a single language is configured, it skips classification and returns that language directly.

Related projects

• ovos-audio-transformer-plugin-bandpass - band-pass filter audio transformer for OpenVoiceOS
• ovos-audio-transformer-plugin-ggwave - data-over-sound audio transformer for OpenVoiceOS

License

Apache-2.0